Abstract: The recognition and communication accuracy of speech system is considered to be decreased under strong noise environment. To solve the problem, a speech enhancement method is presented which is based on adaptive noise estimation. In the new algorithm, the speech signal is divided into speech segments and non-speech segments by endpoint detection, and the noise amplitude spectrums of the two kings segments are estimated adaptively and respectively. This algorithm also improves the spectral subtraction principle formulas according to the research of hypothesis in spectral subtraction which is not common. The experimental result shows that the algorithm in this paper performs better in reducing musical noise, maintaining high clarity and intelligibility, and improving the speech recognition and communication accuracy under strong noise environment than traditional spectral subtraction.

Key words: speech enhancement, spectral subtraction, noise estimation, music noise

摘要： 针对语音系统受外界强噪声干扰而导致识别精度降低以及通信质量受损的问题，提出一种基于自适应噪声估计的语音增强方法。通过端点检测将语音信号分为语音段与非语音段，对这两种情况的噪声幅度谱分别进行自适应估计，并对谱减法中不具有通用性的假设进行研究从而改进原理公式。实验结果表明，相对于传统谱减法，该方法能更好地抑制音乐噪声，并保持较高清晰度和可懂度，提高了强噪声环境下的语音识别精度和通信质量。

关键词: 语音增强, 谱减法, 噪声估计, 音乐噪声